Context Declaration
For Human Readers, AI Assistants, and the Research Community
A Vision Before the Detail
This document describes a framework for making better decisions under deep uncertainty. Before entering the architecture, the vocabulary, or the proof of concept, it is worth pausing on what that means in practice, and what it could mean at scale.
Consider a large dairy processor evaluating whether to electrify its heat supply by 2035. The decision looks, on the surface, like a site-level engineering and finance problem: compare the capital cost of electric boilers against biomass conversion, estimate operating costs, account for the carbon price, choose the pathway with the lower net present value. That framing is not wrong. It is incomplete. The pathway choice propagates consequences to a regional electricity grid that may not have the capacity to absorb the additional load without reinforcement. That reinforcement, if required, takes years to plan, consent, and build. It also draws on the same grid capacity that other industrial users in the region are simultaneously seeking to access. The biomass pathway, meanwhile, draws on a regional feedstock supply that other users are simultaneously competing for. Both pathways are exposed to a carbon price whose trajectory is set by a policy process that is institutionally contested and whose design has changed materially several times in the past decade. And the facility’s decision will be made by a corporate sustainability team, reviewed by a board with both financial and reputational objectives, influenced by a government co-investment programme with its own eligibility criteria and funding cycles, shaped by community expectations about the future of the Southland economy, and ultimately constrained by physical infrastructure that was built for a different industrial era.
None of this complexity is unusual. It is the normal texture of a consequential infrastructure decision in the early twenty-first century. The problem is that the analytical tools most commonly brought to bear on this kind of decision were designed for a simpler version of it: one site, one future, one objective, one model boundary. They are not inadequate because they are unsophisticated. They are inadequate because they are organised around the wrong centre of gravity. They are organised around the system rather than around the decision.
This framework is organised around the decision. That reversal, simple to state and demanding to operationalise, is what the rest of this document develops.
What the Framework Is Attempting
The framework developed here begins from a proposition that is easy to accept in the abstract and surprisingly difficult to implement in practice: the model should develop toward the parameters that influence the decision, not toward the completeness of the system description.
In conventional modelling practice, a system is first described as completely as possible within a boundary defined by physical or technical extent, and the decision is then framed as a query to that model. The model defines what can be asked. In the framework proposed here, the process is reversed. The decision is articulated first: what is being chosen, by whom, with consequences at which scales, and under which dimensions of uncertainty. The model then develops toward the parameters that influence that decision. Physical detail is incorporated where it changes the outcome of the comparison. Institutional conditions are represented where they shift the ranking of alternatives across futures. Social and environmental consequences are included where they are relevant to what the decision-maker must account for, not as afterthoughts appended to a completed technical analysis but as first-class dimensions of the decision frame from the outset.
This means the framework grows. It begins with a minimum viable representation that is honest about its simplifications and explicit about where additional detail would change the decision-relevant comparison. At each development step, it asks: what is the most influential parameter not yet well represented? What constraint, once modelled, would shift the robustness ranking of the alternatives? What future condition, once included, would reveal a vulnerability currently invisible? The answer to these questions, rather than a commitment to physical completeness, determines what gets modelled next.
The consequence is an analytical environment that is at once humble and ambitious. Humble, because it never claims to have modelled everything that matters. Ambitious, because it claims to have modelled exactly what matters most at each stage of development, and to be able to demonstrate that claim through the traceability of its artefacts and the transparency of its decision-relevant comparisons.
There is a further property that makes this ambition practically important. The framework is self-directing in its development. At each stage, the regret sensitivity diagnostics and the scenario discovery analysis of the DMDU orchestration layer identify which uncertain drivers most strongly determine pathway preference and therefore which simplification, once removed, would most change the decision-relevant comparison. The framework lights up the domain that needs to be explored next. It does not promise to eliminate uncertainty. It promises to make decisions more intelligible in its presence, and to reveal, at each step, where the analytical torch should be turned next.
The Broader Vision: What Full Deployment Could Achieve
The Edendale–Southland proof of concept documented in this manuscript is, by the framework’s own declared standards, a first-generation demonstration. It demonstrates that the architecture works, that the site-to-region coupling reveals consequences invisible from within the site boundary, and that the artefact governance chain makes those findings traceable and reproducible. But the Edendale case is one node in a far larger decision landscape, and it is worth being explicit about what the framework could achieve if deployed at greater scale and depth.
At the corporate scale, a single company such as Fonterra operates multiple large processing sites across New Zealand, each with its own pathway decision, each connected to regional grid infrastructure at different GXPs, each drawing on regional biomass and other resource systems with different supply chain characteristics. The pathway decision at Edendale does not exist in isolation from the decisions being made at Canpac, Waitoa, Darfield, or Clandeboye. These decisions interact through shared grid infrastructure, shared biomass markets, and a shared carbon pricing environment. A company-scale deployment of the framework would evaluate all site-level pathway decisions simultaneously, using a shared regional electricity module that represents the aggregate demand pressure on the national transmission system, a shared biomass resource module that represents competition for regional feedstock supply, and a shared policy module that represents the ETS carbon price trajectory and its interaction with the investment timing of all facilities.
Beyond strategic pathway planning, the same architecture supports operational optimisation. Once a pathway has been selected and implemented, the site module can be configured to support day-ahead and week-ahead dispatch optimisation, using real-time operational data from the facility’s control systems translated through the artefact interface. Existing energy management system data, existing engineering assessments, and existing financial models all translate into governed artefacts without requiring the organisation to start from scratch. The strategic and operational layers of the same organisation operate within the same analytical environment, connected through the same backbone: strategic pathway analysis informs which assets are available; operational dispatch analysis demonstrates how those assets perform day to day; and the accumulated operational record informs the next round of strategic reassessment.
The result of a corporate deployment would not be an optimal transition plan, because no such plan can be specified under the conditions of deep uncertainty the framework acknowledges. It would instead be a robust transition strategy: a set of sequencing and staging decisions that remains defensible across many plausible futures, where those futures include not only grid and biomass supply conditions but also the decisions of competing users, the responses of regulators, and the evolution of the corporate sustainability reporting environment.
At the sector and regional scale, the same architecture extends to encompass multiple companies and multiple sectors. The Southland region’s dairy, meat, and wood products processors share the same regional grid, the same biomass resource zone, and the same policy environment. A regional deployment of the framework would represent all major industrial heat users simultaneously, allowing the analytical environment to evaluate not only individual site decisions but the system-level consequences of different regional transition configurations. This is a question that no individual site assessment can answer, that no national model currently addresses at the required spatial and temporal resolution, and that the framework’s architecture is specifically designed to support.
At the national scale, the framework’s architecture is compatible with coupling to the national-scale models that already exist. TIMES-NZ, New Zealand’s national energy system model, produces economy-wide scenario outputs including electricity price trajectories, sectoral fuel demand projections, and national emissions pathways. These outputs represent exactly the kind of exogenous future conditions that belong in a FutureArtefact ensemble. A national-scale deployment would use TIMES-NZ scenario outputs to calibrate the ranges of the uncertain drivers in the ensemble, ensuring that site-level and regional-level pathway comparisons are consistent with plausible national energy trajectories. The initial coupling would be one-directional: TIMES-NZ scenarios inform the ensemble design. The eventual next phase would implement the reverse coupling, allowing the aggregate industrial pathway decisions revealed by the framework to be summarised as demand scenarios and passed back to TIMES-NZ, so that national and regional levels co-evolve iteratively through the Gauss-Seidel coupling architecture described in the framework’s next-phase specification.
At the domain level, the framework’s decision-first architecture is not confined to energy. Water system planners in a drought-exposed catchment, regional transport authorities evaluating infrastructure investment under uncertain demand, land-use planners navigating competing development pressures, and industrial decarbonisation investors in Southeast Asia, the European industrial heartland, and the Gulf petrochemical complex all face the same structural problem: long-lived commitments, interacting infrastructure layers, deep uncertainty about future conditions, and the need for decisions that are accountable to multiple stakeholders with different perspectives. Each of these domains can be instantiated within the framework’s architecture by specifying the relevant decision class, the relevant module structure, and the relevant artefact families. The thin-waist principle ensures that contributions from specialists in each domain can be integrated without requiring those specialists to understand or maintain the full framework.
At the stakeholder accountability level, the framework’s artefact governance architecture creates something genuinely new in planning practice: a traceable record of how a decision-relevant finding was produced, from the raw input data and configuration parameters through the module runs and acceptance gates to the regret and robustness metrics that informed the decision. This traceability is not merely a technical property. It is an institutional one. A decision made on the basis of framework analysis can be audited: a regulator, a community representative, or a future analyst can follow the lineage from the finding back to the assumptions and ask, precisely and specifically, whether the assumptions were appropriate. It becomes possible to say, with specificity: under these conditions about grid headroom, biomass supply, and carbon price trajectory, the electrification pathway generates lower regret than the biomass pathway in 41 of 64 plausible futures, and here is the complete audit trail of the analysis that produced that finding. That is a qualitatively different foundation for public accountability than the outputs of a black-box optimisation.
This accountability dimension extends, ultimately, to questions of energy justice and democratic governance. When long-horizon infrastructure decisions are made on the basis of analyses that are traceable, multi-perspective, and open to community interrogation, the governance of those decisions becomes more democratic in a substantive rather than procedural sense. Communities can participate not merely by attending consultation meetings but by engaging with the analytical basis of the decisions that will shape their regional economies for decades. The framework’s architecture makes that engagement possible in a governed, productive, and cumulatively valuable way.
The Architecture That Makes This Possible
None of the above requires building a model of everything. It requires building an architecture that can accommodate heterogeneous analytical components, connect them through governed interfaces, evaluate their combined outputs under structured uncertainty, and grow progressively toward the parameters that most influence the decisions at hand.
The thin-waist artefact structure is the mechanism that makes this possible. It does not require that all components be simple. The site dispatch module can be a rule-based proportional allocator in the first generation, an LP-based scheduler in the second, and an OpenModelica thermal network simulation in the third. The regional electricity module can be a stylised screening model in the first generation, a full PyPSA network optimisation in the second, and a trained ML surrogate of that optimisation in the third. The biomass resource module can be a scalar multiplier in the first generation and a spatially explicit logistics model calibrated to satellite-derived forest inventory data in the second. The macroeconomic context module can be a TIMES-NZ scenario parameter table in the first generation and a coupled bi-directional model exchange in a later generation. At every generation, what crosses the module boundaries is the same: a schema-conforming artefact with declared provenance and a validated integrity record. The rest of the framework need not know which generation of implementation produced it.
The decision-first principle governs the order in which these developments are pursued. The framework does not develop toward greater complexity in general. It develops toward greater decision relevance specifically. The question asked at each development step is: what is the most consequential simplification currently in the model, in the sense that removing it would most change the decision-relevant comparison? The answer to that question, produced by the regret sensitivity diagnostics and the scenario discovery analysis of the DMDU orchestration layer, is the framework’s own contribution to determining what should be developed next.
Who This Is For
The framework described in this document is addressed to at least six audiences, and it asks different things from each. What it offers all of them is a version of the same thing: a way of organising analytical complexity that keeps the decision visible, keeps the assumptions traceable, and keeps the path toward better analysis open as knowledge grows.
It asks researchers to engage with a methodological argument that cuts across several established traditions without belonging entirely to any of them.
In complex planning environments, the quality of decision support depends more on analytical architecture than on model sophistication. A more powerful solver embedded in a poorly framed decision environment produces better answers to the wrong question. A well-governed, decision-centred architecture produces answers that are less precise in a narrow mathematical sense but more defensible in the sense that actually matters: they are traceable to their assumptions, comparable across alternative futures, and revisable as knowledge improves.
This is a substantive intellectual contribution to the DMDU tradition rather than a repackaging of existing ideas. The RDM community, represented in the work of Lempert, Popper, Bankes, Walker, Marchau, Kwakkel, and Haasnoot, has established the evaluative standards of regret, robustness, and satisficing and has developed the computational methods of scenario discovery and exploratory modelling. What it has not systematically addressed is the architectural question of how modular, multi-scale analytical environments should be organised so that DMDU-style exploration can operate across institutional and physical boundaries simultaneously, with governed artefact exchange that preserves traceability across all scales. The thin-waist artefact architecture, the decision-first boundary principle, and the progressive-refinement philosophy are contributions to that open question.
Researchers working in adjacent traditions will find points of engagement throughout. Those working in integrated assessment modelling will find the framework’s treatment of scale heterogeneity and the site-to-region coupling directly relevant to the persistent challenge of connecting bottom-up engineering detail to top-down economic consistency without sacrificing either. Those working in socio-technical systems theory will find the framework’s treatment of multi-actor perspectives and system-level regret directly relevant to the challenge of representing institutional complexity in planning models in a form that is analytically tractable rather than merely acknowledged. Those working in the philosophy of science will find the framework’s post-positivist epistemological commitments, developed in Sub-Module SM-1.1-A, a substantive engagement with questions about what models are for and what counts as analytical legitimacy under deep uncertainty.
It asks energy system modellers to consider how the tools they already use and trust can contribute to something larger than any one of them can achieve alone.
The open-source energy modelling ecosystem is mature and growing. PyPSA provides network-aware electricity system optimisation with an active international development community and a growing library of regional and national model implementations. Linopy, the algebraic modelling layer underlying PyPSA’s latest architecture, makes large network problems tractable with a clean Python interface. OpenModelica provides equation-based physical simulation of thermal, electrical, and fluid systems with modular, reusable component libraries suited to site-level industrial process modelling. Calliope and OSeMOSYS provide flexible, transparent energy system frameworks with strong open-data commitments and active academic communities. EnergyPLAN provides integrated energy systems simulation at the regional and national scale. TIMES, MESSAGE-ix, and the broader IAM family provide national and global scenario analysis capabilities with decades of institutional application and established policy uptake. Each of these tools is internally sophisticated. None of them, on its own, addresses the full analytical chain from a site-level operational decision through a regional infrastructure constraint to a national energy trajectory, under deep uncertainty, in a form that is simultaneously governed, traceable, and accountable to multiple stakeholders with different perspectives.
The framework proposed here is not a replacement for any of these tools. It is an architecture within which they can contribute to that full analytical chain. PyPSA belongs in the Regional Module, where its network-aware dispatch optimisation produces the adequacy signals and dual-variable price indicators that the site-level pathway evaluation requires. OpenModelica belongs in the Facility Module, where its thermal network simulation produces the high-fidelity operational detail that a scheduling-grade dispatch model needs. TIMES-NZ belongs in the FutureArtefact ensemble calibration layer, where its national scenario outputs provide the macroeconomic and policy boundary conditions within which regional and site-level futures are situated. The framework tells each tool what it is responsible for producing, in what form, and with what provenance. It does not tell each tool how to produce it. The internal sophistication of each tool is liberated rather than constrained by the architecture.
The practical value of the surrogate development pathway deserves specific emphasis. A full PyPSA regional network model for the Southland context is computationally expensive to run. Running it inside a DMDU ensemble of several hundred or thousand futures is currently impractical. Training a decision-ranked ML surrogate of that model, following the validation protocol described in Sub-Module SM-2.3-A, makes the ensemble expansion tractable without sacrificing the decision-ranking fidelity that the comparison requires. The surrogate does not replace PyPSA; it emulates PyPSA’s input-output behaviour at the level of the decision-relevant outputs the framework requires, and the full PyPSA model remains available for validation and for cases near regime boundaries where the surrogate’s confidence score falls below the declared threshold. The same surrogate architecture applies to a spatially explicit biomass logistics model, to a thermal process simulation, and to a simplified coupling with macroeconomic scenario generators such as TIMES-NZ. At every generation of implementation, what crosses the module boundary is the same schema-conforming artefact. The analytical chain does not need to know which generation produced it.
For modellers working with the IEA Project BlueSky interoperability initiative, the framework offers a complementary governance layer. BlueSky’s ambition of standardised interfaces between heterogeneous modelling components is architecturally aligned with the thin-waist principle developed here. The additional contribution of the present framework is artefact governance: the schema conformance, provenance declaration, validation gating, and lineage tracking that make modelling contributions not only interoperable but auditable across the full analytical chain.
It asks corporate decision-makers to recognise that the analytical environment they need for long-horizon investment decisions under deep uncertainty already exists in fragmentary form inside their organisations, and that the framework’s contribution is to connect those fragments in a governed, coherent way that creates value they cannot currently access.
Large industrial companies managing long-lived asset portfolios in a decarbonising economy are not operating without analytical support. They have energy management systems that record real-time and historical operational data at hourly or sub-hourly resolution. They have financial models that project operating costs under different fuel price and carbon price scenarios. They have engineering assessments that characterise the technical performance of alternative technology pathways. They have sustainability reporting systems that aggregate emissions data across facilities. They have grid connection assessments that characterise the available hosting capacity at their points of connection. What they typically lack is an analytical architecture that connects these data sources through governed interfaces, evaluates the pathways they reveal under structured uncertainty across multiple futures, and produces decision-relevant metrics that are simultaneously precise enough to be actionable and honest enough to be defensible under external scrutiny.
The framework is designed to receive the data that already exists within industrial organisations and translate it into governed artefacts. An existing energy management system’s operational data becomes the basis for a calibrated DemandPack rather than a synthetic one. An existing engineering assessment of boiler replacement costs becomes the capital cost parameter in the site module’s configuration artefact. An existing financial model’s fuel price and carbon price projections become the calibration basis for the FutureArtefact ensemble’s uncertain driver ranges. An existing grid connection assessment becomes the reference capacity estimate around which the headroom multiplier is scaled. None of this requires the organisation to start from scratch. It requires the existing analytical work to be translated into the framework’s artefact format, which is the precondition for connecting it to the regional and policy layers that the site-level analysis alone cannot address.
The commercial advantage this offers is specific and demonstrable. A corporate investment decision made on the basis of a governed, traceable, multi-layer robustness analysis is a qualitatively different document from one made on the basis of a point-estimate techno-economic assessment. It can demonstrate to the board that the pathway recommendation remains defensible under a declared range of future conditions rather than only under central-case assumptions. It can demonstrate to investors that the capital allocation decision has been stress-tested against the scenarios that ESG-focused investment frameworks now require. It can demonstrate to regulators that the company has evaluated the system-level consequences of its pathway choice and has not simply optimised within its own boundary at the expense of shared infrastructure resources.
For a company managing multiple sites, the multi-site extension offers a further commercial advantage: the ability to evaluate interaction effects between sites explicitly. The aggregate electricity demand of multiple electrification pathway commitments across a South Island portfolio is not the sum of individual site assessments. It is a system-level demand profile that hits multiple GXPs simultaneously, competes for reinforcement investment prioritisation by Transpower, and creates aggregate biomass demand that may constrain the supply available to any individual site. The corporate-scale deployment of the framework makes these interactions visible and includes them in the pathway comparison. A corporate sustainability officer who can demonstrate that a portfolio transition strategy has been evaluated against system-level interaction effects occupies a significantly stronger position in negotiations with energy regulators, network operators, and co-investment agencies than one who cannot.
It asks policy-makers and regulators to recognise that the system-level incentive misalignments revealed by the framework’s multi-perspective evaluation are not academic abstractions but consequences of real, specific architectural features of the current policy and pricing environment, features that can be identified precisely, attributed to specific cost components and pricing arrangements, and addressed through targeted policy design.
The framework’s most important policy finding to date is the divergence between site-perspective and system-perspective cost assessments for the electrification pathway under GXP-constrained futures. Under current network pricing arrangements, the cost of grid reinforcement triggered by a new large industrial load may not be directly charged to the load that triggered it, creating a systematic incentive for individual site operators to pursue electrification pathways whose private cost appears competitive even when the full-system cost is significantly higher. This misalignment is not visible in any single-site techno-economic assessment. It is only visible when the site’s incremental electricity demand is evaluated against the regional grid system’s actual hosting capacity constraints, and when the cost of the resulting reinforcement requirement is attributed from both the private and public perspectives simultaneously. The framework makes this attribution explicit, and it attributes it to specific futures and specific cost components, so that the policy question becomes precise: under which conditions do the misalignment occur, how large is it, and which policy instrument most directly addresses it?
For EECA, the framework offers a complement to the RETA screening methodology that addresses exactly the limitation that RETA’s current approach cannot reach: the time-resolved, uncertainty-aware, multi-perspective evaluation of individual site pathway decisions against regional infrastructure constraints. RETA identifies where the decarbonisation opportunities are and what infrastructure would be needed to realise them. The framework evaluates whether those opportunities remain robust across the futures in which the infrastructure may or may not materialise on the required timeline. The two approaches are designed for each other.
For the Climate Change Commission, the framework offers a tool for evaluating whether the sectoral emissions reduction targets in the emissions reduction plan are achievable under the specific infrastructure and supply chain conditions that determine feasibility at the site level. A national emissions reduction plan that requires large-scale industrial electrification without representing the regional grid infrastructure constraints that determine feasibility on the required timeline may look achievable in a national model and be infeasible on the ground. The framework’s site-to-region coupling makes that gap analytically visible and attributable.
For the Electricity Authority and Transpower, the framework offers a structured approach to the aggregate demand forecasting challenge that large-scale industrial electrification creates. Multiple industrial users electrifying simultaneously in the same region creates aggregate GXP demand that is difficult to predict from individual site assessments alone. A framework-based regional demand assessment, aggregating IncrementalElectricityPacks from multiple facility modules through a shared regional module, produces an aggregate demand forecast that is grounded in individual facility operational characteristics, evaluated across a structured future ensemble, and expressed in the time-resolved format that network investment planning requires.
It asks AI assistants and the research community to engage with this document not as a completed account of a finished framework but as the first formal specification of a knowledge environment that is designed to grow through exactly the kind of engagement that AI-assisted, community-driven research can provide.
The framework’s modular architecture is the blueprint for its own extension. A domain specialist in water infrastructure planning who reads this document and recognises the decision structure they are working with can contribute a domain instantiation by specifying the relevant module classes, artefact families, and uncertain drivers in terms of the framework’s interface contract. A PyPSA developer who reads the regional module specification in Sub-Module SM-6.6-E can implement that module and contribute it to the community through the process described in the Context Declaration. A researcher in the DMDU community who identifies a new robustness metric or a new scenario discovery algorithm that improves the evaluation layer can propose its integration as a methodological extension. Each of these contributions enters a governed environment: it is assessed for compatibility with the thin-waist architecture and the decision-first principle, reviewed through the editorial process described in the Context Declaration, and if accepted, formally incorporated with full provenance.
The AI-assisted engagement pathway deserves specific mention. A practitioner who uploads this document to a capable AI assistant and asks questions drawn from the audience-specific catalogues is using the document in exactly the way it was designed to be used: as a structured knowledge base that supports precise, framework-grounded engagement with the specific question at hand. The AI assistant is not being asked to summarise the document. It is being asked to reason about the framework with the precision that the framework’s own vocabulary supports, to distinguish between what has been implemented and what has been specified, to help the user navigate from their specific question to the part of the framework most relevant to addressing it, and to acknowledge explicitly when a question concerns something the current framework does not yet represent.
The framework does not need to be complete to be useful. It needs to be honest about what it does and does not yet represent, to provide a clear path toward representing what it does not, and to create the conditions under which the research community and the communities of practice most directly affected by the decisions the framework supports can contribute to that path in a governed, accountable, and cumulatively productive way. This document is the beginning of that path.
A Note on How the AI Assistant Should Use This Document
When answering any question using this document, the AI assistant should ground answers in the framework’s specific terms and concepts rather than in general knowledge about energy systems or decision theory that may differ from how this framework uses those terms. It should acknowledge explicitly when a question asks about something the current framework does not yet implement and explain what would be required to address it, consistent with the progressive-refinement philosophy. It should distinguish between what the Edendale proof of concept has demonstrated, what has been specified as next-phase development, and what remains a longer-range vision.
The assistant should treat the Node Declaration Tables at the front of each module as the authoritative map of what each module contributes and what it assumes. It should use the numbered propositions in Module 7 §7.6 as the distilled claim set when asked about the framework’s intellectual contribution. It should treat the decision-first boundary principle as the conceptual anchor that connects every architectural choice to the framework’s foundational commitment. When a question concerns a domain outside process heat, the assistant should use the framework’s architecture as the basis for reasoning about how that domain would be instantiated, specifying the relevant decision class, the relevant module structure, and the relevant artefact families in terms the framework’s interface contract supports.
What This Document Contains
The document is organised across eight modules preceded by the Context Declaration, Preamble, and Navigation Guide. Every module opens with a Node Declaration Table specifying what it assumes, what it contributes, whether it can be read independently, and what it passes forward. Module 0 provides the framework map and orientation. Modules 1 and 2 establish conceptual and methodological foundations. Module 3 develops the architecture. Modules 4 and 5 instantiate it in the energy domain and the New Zealand context. Module 6 documents the full Edendale–Southland proof of concept. Module 7 synthesises, extends, and closes the argument.
Module 0: Orientation provides the framework map, the decision-first boundary principle, the seven functional layers, the current implementation status, and the navigational logic for the rest of the document.
Module 1 establishes the conceptual and methodological foundations: why decision-first boundary setting is the appropriate response to long-horizon planning under deep uncertainty, what analytical grammar is needed to pursue it rigorously, and what tools are available to generate decision-relevant consequences.
Module 3 develops the architecture: modular decomposition, thin-waist artefact exchange, and the governed data backbone that makes the environment persistent, traceable, and extensible.
Modules 4, 5, and 6 instantiate the architecture: first in the domain of coupled energy systems and industrial process heat, then in the specific New Zealand conditions that motivate the proof of concept, and finally in the full documented Edendale–Southland proof of concept with its implemented pipeline, its artefact integrity record, and its results.
Module 7 synthesises extending the framework beyond its current instantiation, stating its limits honestly, and presenting the eleven propositions that constitute its intellectual claim.
Current Implementation Status
| Component | Status | Significance |
|---|---|---|
| DemandPack construction (synthetic, RETA-calibrated) | Implemented | Foundational demand artefact for all epochs |
| Proportional and optimal-subset dispatch | Implemented | Site-level pathway evaluation with cost and emissions accounting |
| SignalsPack generator (Edendale GXP, schema v0.1.0) | Implemented | Frozen interface contract with SHA256 hash integrity |
| GXP screening module (stylised) | Implemented | First-generation regional adequacy assessment |
| Grid RDM evaluation (21 to 100 futures) | Implemented | Robustness screening across five uncertain driver dimensions |
| Site decision robustness overlay | Implemented | Site-cost uncertainty separated from grid-side uncertainty |
| Pathway comparison (2035_EB vs 2035_BB) | Implemented | Paired-futures regret and robustness diagnostics |
| DuckDB and Parquet analytical backbone | Specified | Governed, append-only, queryable persistence layer |
| PyPSA regional electricity module | Specified | Full network optimisation replacing stylised screen |
| OpenModelica site truth model | Specified | Equation-based physical simulation replacing proportional dispatch |
| Surrogate-accelerated ensemble | Specified | ML emulator enabling ensemble expansion to thousands of futures |
| TIMES-NZ scenario coupling | Vision | National-regional consistency in FutureArtefact ensemble calibration |
| Corporate-scale multi-site deployment | Vision | Portfolio transition strategy across multiple facilities |
| Multi-domain extension | Vision | Water, transport, and land-use instantiations through community contribution |
| Natural-language query interface | Vision | Accountability-grade interrogation of the governed artefact store |
Key Concepts Defined Here
The following terms are used throughout this document in the precise senses defined below. Full definitions for all framework terms are in the Glossary.
Decision-first boundary principle. Analytical boundaries are drawn by the decision context — what must remain visible for the comparison to be valid — not by the physical extent of the system.
Artefact. A governed analytical output: schema-conforming, provenance-carrying, and validation-gated before it enters the comparison chain.
Thin waist. What crosses module boundaries must be narrow, explicit, and stable. Module internals may evolve freely provided the interface contract is preserved.
Deep uncertainty. Conditions where probabilities over futures are structurally unavailable or contested, requiring robustness evaluation rather than expected-value optimisation.
Regret. The difference between a chosen strategy’s outcome and the best available outcome under the same future.
Robustness. A strategy’s property of performing acceptably across many plausible futures, not only under the central case.
Satisficing. Meeting declared thresholds of acceptability across futures rather than nominal optimisation against any single one.
Progressive refinement. The development philosophy: add complexity where regret diagnostics reveal it would most change the decision-relevant comparison.
GXP (Grid Exit Point). The metered boundary between the national transmission grid and a regional distribution network — the primary infrastructure constraint for industrial electrification in the NZ context.
One-pass coupling. The current PoC limitation: modules run in sequence without iterative feedback between site dispatch and regional screening.
Community Extension
The document is released as a living resource. The framework’s modular architecture is designed to accommodate extension without disruption: a new domain instantiation adds a new Part-equivalent appendix; a new artefact family adds a new schema entry; a new uncertain driver adds a field to the FutureArtefact; a new regional application adds a new proof-of-concept chapter.
Contributions are welcomed in four categories. Domain instantiations apply the framework to domains beyond industrial process heat. Each instantiation should specify the relevant decision class, module structure, and artefact families in terms of the framework’s interface contract. Module implementations provide alternative implementations of any module satisfying the same interface contract. Methodological extensions extend the evaluation layer with new robustness metrics, scenario discovery approaches, or adaptive pathway formulations. Corrections and critiques identify errors or propose alternative approaches.
Contributions may be submitted through the GitHub repository associated with this document, using the structured templates provided in the GitHub Discussions space. The editorial process assesses compatibility with the thin-waist architecture and the decision-first principle before formal incorporation.